经 AI Skill Hub 精选评估,跨境卖家自动化 获评「强烈推荐」。这款Agent工作流在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 8.0 分,适合有一定技术背景的用户使用。
跨境卖家自动化 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
跨境卖家自动化 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install vibe-seller
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install vibe-seller
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/zpoint/vibe-seller
cd vibe-seller
pip install -e .
# 验证安装
python -c "import vibe_seller; print('安装成功')"
# 命令行使用
vibe-seller --help
# 基本用法
vibe-seller input_file -o output_file
# Python 代码中调用
import vibe_seller
# 示例
result = vibe_seller.process("input")
print(result)
# vibe-seller 配置文件示例(config.yml) app: name: "vibe-seller" debug: false log_level: "INFO" # 运行时指定配置文件 vibe-seller --config config.yml # 或通过环境变量配置 export VIBE_SELLER_API_KEY="your-key" export VIBE_SELLER_OUTPUT_DIR="./output"
<p align="center"> <b>Multi-store browser-automation framework for cross-border sellers.</b><br/> Runs on macOS, Linux, or Windows (native installer). Drives real browsers — Ziniao for fingerprint isolation, plain Chrome otherwise. Bring your own LLM API key — Claude, DeepSeek, GLM, Qwen, Kimi, MiniMax all work. </p>
<p align="center"> <a href="README.md">English</a> · <a href="README.zh.md">中文</a> </p>
<p align="center"><img width="1200" height="338" alt="combined_small" src="docs/images/combined_small.gif" /></p>
<img width="2779" height="1737" alt="1_combined" src="docs/images/1_combined.png" />
<img width="1989" height="1490" alt="2_combined" src="docs/images/2_combined.png" />
---
Run VibeSeller-Setup.exe (or the PowerShell one-liner from Install). It starts the server for you — check Open Vibe Seller now at the end of the wizard, or open <http://localhost:7777> yourself.
Copy this entire line, paste into the terminal, hit return:
curl -sSL https://raw.githubusercontent.com/zpoint/vibe-seller/main/install.sh | bash
Takes a few minutes (downloads the Python toolchain, installs Vibe Seller, fetches Chromium). When the output ends with Vibe Seller installed! you're set.
Stuck or hitting an error? Drop the repo URL <https://github.com/zpoint/vibe-seller> into any coding agent (Claude Code, Codex, opencode, Cursor) and say "read the README and install this." The agent will work through it for you.
Copy this entire line, paste into the terminal, hit return:
curl -sSL https://raw.githubusercontent.com/zpoint/vibe-seller/main/install.sh | bash
Takes a few minutes (downloads the Python toolchain, installs Vibe Seller, fetches Chromium). When the output ends with Vibe Seller installed! you're set.
Stuck or hitting an error? Drop the repo URL <https://github.com/zpoint/vibe-seller> into any coding agent (Claude Code, Codex, opencode, Cursor) and say "read the README and install this." The agent will work through it for you.
<details open> <summary><b>🪟 Windows — native installer (recommended)</b></summary>
No WSL, no Python setup. In PowerShell:
irm https://raw.githubusercontent.com/zpoint/vibe-seller/main/installer/windows/install.ps1 | iex
Or download VibeSeller-Setup.exe from the latest release and run it.
</details>
<details open> <summary><b>🍎 macOS / 🐧 Linux</b></summary>
curl -sSL https://raw.githubusercontent.com/zpoint/vibe-seller/main/install.sh | bash
vibe-seller start
Open <http://localhost:7777>. Upgrade: vibe-seller upgrade. Uninstall: uv tool uninstall vibe-seller.
<sub>Prerequisites (auto-handled by <code>install.sh</code>): Python 3.11+ via <a href="https://docs.astral.sh/uv/"><code>uv</code></a>, Node.js 22+.</sub>
</details>
<details> <summary>🪟 Windows via WSL2 (advanced)</summary>
Prefer the native installer above. The WSL2 path runs the server inside Ubuntu/WSL with the browser on the Windows host — it needs mirrored networking (Windows 11+). Full setup lives in the developer guide.
```bash
Want the agent to do daily email triage and push critical items to a group chat? Settings → Integrations:
- Email: enter the mailbox address for each store; the IMAP server fields auto-populate (gmail, outlook, self-hosted all supported), so all you fill in is the secret (an app password usually). Not sure how to generate one? Ask any AI with the name of your email provider. Agent scans daily afterwards. - WeCom (企业微信): set up a group bot, paste the webhook URL. High-priority anomalies post automatically. - TickTick / Google Workspace: connect here if you want results synced into a calendar or Doc.
Skipping is fine — come back after your first task.
Want the agent to do daily email triage and push critical items to a group chat? Settings → Integrations:
- Email: enter the mailbox address for each store; the IMAP server fields auto-populate (gmail, outlook, self-hosted all supported), so all you fill in is the secret (an app password usually). Not sure how to generate one? Ask any AI with the name of your email provider. Agent scans daily afterwards. - WeCom (企业微信): set up a group bot, paste the webhook URL. High-priority anomalies post automatically. - TickTick / Google Workspace: connect here if you want results synced into a calendar or Doc.
Skipping is fine — come back after your first task.
Beyond the browser, the agent can call these external services directly (hook up permissions in First run — no code):
| Service | What it does |
|---|---|
| **Email** (IMAP/SMTP) | Read your bound store inboxes, classify, forward, draft replies |
| **WeCom** (企业微信) | Push important events, cross-group notifications, forward buyer inquiries |
| **TickTick / Dida365** | Auto-create todos (follow-ups, order tracking, periodic checks) |
| **Google Workspace** | Gmail, Drive, Docs, Sheets, Slides, Calendar — read, write, edit |
A single task can mix browser + service calls. e.g. "Find all slow-moving SKUs across stores" → browser pulls inventory → generates a Google Sheet → posts a WeCom summary → drops a "price review next week" todo into TickTick.
<details> <summary><b>How much does it cost?</b></summary>
Vibe Seller is free and open source. You pay your LLM provider directly. Rough scale: one ad-tuning run ≈ ¥1 in DeepSeek tokens.
</details>
<details> <summary><b>Which LLM should I pick?</b></summary>
Different tasks need different model strength. Match the model to the task:
| Task type | Recommended |
|---|---|
| **Simple** — email classification, fixed-flow data entry, structured extraction | MiniMax, Kimi, GLM, or Qwen — pick whatever's cheapest |
| **Complex** — exploring a new platform (no built-in Skill), ad optimization, inventory strategy, long multi-page flows | **Claude or DeepSeek. Don't use the cheap models here.** |
Why? Complex tasks ask the agent to read the Skill, judge page state, call tools, and take notes — all in the same context window. Cheap models (the low-tier MiniMax tiers especially) commonly:
One often-overlooked factor: context window size. Claude and DeepSeek v4 ship with 1M context windows; most other models sit around 200k. On a long task (multi-page tables, cross-flow conversation history, accumulated notes), 200k fills up fast — the model starts forgetting earlier decisions, dropping details, or breaking outright. 1M handles the same task with room to spare.
For simple tasks, model choice barely matters; save tokens with a cheap one. For hard tasks, paying for a SOTA model isn't a luxury — it's the cheaper option once you count failed runs.
</details>
<details> <summary><b>How does Ziniao integration actually work? Developer mode vs normal mode / credentials needed / where they're stored</b></summary>
Ziniao runs in exactly one mode at a time:
- Normal mode — what you get when you double-click the Ziniao icon. UI-only, designed for manual daily work. No remote control. - Developer mode (WebDriver) — same Ziniao install, launched with extra flags so it exposes an HTTP API on 127.0.0.1:16851. This is the only mode Vibe Seller (or any external tool) can drive.
The two modes share one Ziniao process, so they can't run side by side. Switching requires restarting Ziniao.
- macOS / native Windows: Vibe Seller handles the swap itself — kills any normal-mode instance and relaunches with --run_type=web_driver when you open a store. Same automatic handling on both platforms, nothing for you to do. - Windows (WSL): WSL can't restart a Windows-side app across the VM boundary, so the Windows side runs a one-line .bat launcher (downloadable from `Settings → Stores → Download launcher`) that does the kill + relaunch. Double-click it once; Vibe Seller picks the developer-mode session up via Refresh.
Credentials needed — Settings → Stores → Add Ziniao account, three fields:
| Field | What it is |
|---|---|
| **Company** (公司名称) | Your Ziniao enterprise company identifier |
| **Username** (企业用户名) | Sub-account username under that company |
| **Password** | That sub-account's password |
Same credentials you use to log into Ziniao — Vibe Seller passes them to the developer-mode HTTP API to list your browsers and start sessions.
The BOSS account also needs to enable developer mode for your company on the Ziniao Open Platform (one-time admin setup, see <https://open.ziniao.com/docSupport?docId=99>). Without it, Ziniao rejects the login (statusCode: -10003 — Ziniao's own error for "developer mode isn't enabled for this company").
Where credentials are stored — all at ~/.vibe-seller/data/vibe_seller.db (local SQLite). No upload, no telemetry path carries credentials out.
- Password is encrypted at rest with a key derived from your local install, and never returned by any HTTP endpoint — decrypted in memory only when Vibe Seller calls the Ziniao HTTP API. - Company and username are plain strings — identifiers, not secrets (same shape as an email address).
Asking for the password is the necessary trade-off: developer-mode auth requires it, no OAuth / token flow exists on the Ziniao side today. If that's a deal-breaker, use plain Chrome instead (Settings → Stores → Add Chrome store). Contributors curious about the exact encryption implementation can check docs/dev-guide.md.
</details>
<details> <summary><b>Ziniao says "无权限通过 webdriver 登录" when I bind a store. How do I fix it?</b></summary>
The BOSS (admin) account hasn't enabled WebDriver login for your sub-account yet. Follow Ziniao's official guide and tick the WebDriver-login option in the Open Platform console:
<img width="1857" height="903" alt="Ziniao Open Platform — enable WebDriver login" src="https://github.com/user-attachments/assets/c6066eff-f877-470e-a061-6abd73b6d2f8" />
The admin (BOSS) account has to do this once. After it's saved, your sub-account can log in via WebDriver and Vibe Seller will bind the store on the next refresh.
</details>
<details> <summary><b>Ziniao says "检测到您在新终端登录" and won't let me bind the store. How do I fix it?</b></summary>
This is Ziniao's new-device security check, not a Vibe Seller bug. The first time a given machine logs into an account, Ziniao requires a one-time manual approval:
1. Open Ziniao on this machine and log in manually (normal mode, not WebDriver) with the same sub-account credentials. 2. Ziniao will show a new-device prompt — submit the approval request. 3. Have the admin (BOSS) account approve the device in the Ziniao backend.
After that one-time approval the device is trusted, and Vibe Seller's WebDriver login goes through on the next refresh.
</details>
<details> <summary><b>Can I run it on a headless server?</b></summary>
Yes — you'll just need to solve the browser/GUI part yourself. Vibe Seller itself runs fine on a headless server; Ziniao (and the Chrome backend, if you want to watch it work) needs somewhere to render into — a virtual display (Xvfb), a cloud desktop VM, or VNC into a real GUI session all work. The simpler common setup, though, is a small always-on desktop machine — a Mac mini, a mini-PC, a spare laptop — that the team connects to over LAN.
</details>
<details> <summary><b>What happens to my store data?</b></summary>
Everything — database, screenshots, logs — stays on your disk under ~/.vibe-seller/. The only outbound traffic is the prompts you send to your LLM provider (which include whatever page snippets the agent reads).
</details>
<details> <summary><b>Where does the agent's "memory" live?</b></summary>
Per-store knowledge accumulates in ~/.vibe-seller/stores/<slug>/ as plain markdown — you can read it, edit it, copy it between machines. Skills are markdown too. Nothing is locked inside a binary blob.
</details>
项目结构清晰,代码质量较高,适合跨境卖家自动化运营
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
AI Skill Hub 点评:跨境卖家自动化 的核心功能完整,质量优秀。对于自动化工程师和运维人员来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | vibe-seller |
| 原始描述 | 开源AI工作流:Multi-store browser-automation framework for cross-border sellers (Ziniao + Chro。⭐11 · Python |
| Topics | 自动化跨境卖家Python |
| GitHub | https://github.com/zpoint/vibe-seller |
| License | Apache-2.0 |
| 语言 | Python |
收录时间:2026-07-02 · 更新时间:2026-07-02 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端